RINA - Démonstrateur d’utilisation de l’intelligence artificielle pour la gestion opérationnelle des risques naturels gravitaires géologiques | FEREC 2020

Actual 0 1 Predicted 0 863 96 1 81 50 Figure 32: Confusion matrix representing the result of predictions based on the Bagging tree classifier method for the train set. Bagging tree classifier Operational accuracy 0.84 0.84 balanced accuracy 0.63 0.60 precision 0.38 0.35 recall 0.34 0.27 F1 0.34 0.30 Table 6: Indices calculated based on the confusion matrix to quantify performance of the bagging tree classifier, for the test and train set, and the operation model. The model produced by the bagging method has superior results for all the indices calculated. It is capable of identifying 3 additional rock falls while increasing the precision of predictions. On the other hand, even though it increases the precision, as it predicts more rock falls it also increases the days the road is closed and an argument could be made that the operational model gives enough safety to drivers while maximizing profit. 5.4. Neural Networks Dense Neural Networks Various models were created by Courteille [Courteille, 2021] to examine the capabilities of NN to predict the rock falls based on the rain. The same training and testing set were used, but the input data included not only the rain of the previous 5 days but instead the rain of the previous 10 days. A DNN was created with one hidden dense layer with relu activation and a final dense layer with softmax activation. Weighing samples by inverse of their occuring frequencies <- unbalanced dataset was also done. The resulting confusion matrix is represented by the Figure 33 41